Model

GPT-5.6 Luna

CLOSED

OpenAI · GPT-5.6 family · released Jul 9, 2026

Cheapest/fastest tier of the three-tier GPT-5.6 family (GA July 9, 2026 alongside Sol and Terra); OpenAI positions the tiers as 'durable capability tiers that can advance on their own cadence' — Luna targets cost-sensitive, high-volume workloads. Same 1.05M context, 128K max output and Feb 2026 knowledge cutoff as Sol. Price cut 80% on July 30, 2026 (launch $1/$6 → $0.20/$1.20 per MTok), taking it from a fifth to a twenty-fifth of Sol's price. Distinct model tier, not a reasoning-effort configuration of Sol (each tier separately exposes its own effort settings).

ReasoningCodingVisionFunction callingTool useAgentic

2534.3

Elo · rank #37

Parameters

Undisclosed

Active params

—

Context

1.05M tokens

Architecture

Undisclosed architecture (presumed dense transformer); fastest/cheapest 'durable capability tier' of the GPT-5.6 Sol/Terra/Luna family, built for cost-sensitive high-volume workloads

License

Proprietary

Languages

—

API price (in/out)

$0.2 / $1.2

Modalities

text · vision

Benchmark results

Bar shows position within the tracked field; marker = field best

Agents' Last ExamAgents50.3%#5

best: GPT-6 Astra · 59.3%

ARC-AGI-1Reasoning88.0%#17

best: Claude Fable 5 · 98.5%

ARC-AGI-2Reasoning59.5%#23

best: GPT-6 Astra · 95.0%

ARC-AGI-3Reasoning0.2%#10

best: GPT-6 Astra · 62.7%

Arena EloHuman preference1452#39

best: Claude Fable 5 · 1504

BrowseCompAgents83.3%#16

best: Kimi K3 · 91.2%

CyberGymCoding77.9%#10

best: MiMo-V2.6-Flash · 95.1%

FrontierBenchAgents14.3%#8

best: Claude Opus 5 · 43.3%

FrontierMath (Tiers 1-3)Math78.6%#3

best: GPT-5.6 Sol · 89.0%

FrontierMath (Tier 4)Math58.5%#3

best: GPT-5.6 Sol · 83.0%

GDPval-AAAgents1591.8#15

best: Claude Fable 5 · 1932

GPQA DiamondReasoning92.3%#20

best: GPT-6 Astra · 96.0%

MMMU-ProVision78.4%#19

best: Claude Opus 4.7 · 85.5%

SWE-bench ProCoding62.7%#13

best: Claude Opus 5.5 · 89.9%

SWE-bench VerifiedCoding93.0%#8

best: Claude Opus 5 · 97.0%

τ³-BankingAgents27.2%#5

best: Step 5 Preview · 42.5%

Terminal-Bench 2.0Coding84.7%#13

best: DeepSeek-V4.1-Flash · 90.6%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

$0.2

Output / M tok

$1.2

GPT-5.6 family

Elo progression across releases

GPT-5.6 LunaJul 20262534.3 GPT-5.6 SolJul 20262829.1 GPT-5.6 TerraJul 20262638.6 GPT-5.6 CyberAug 20261000.0

Resources

DOCSdevelopers.openai.com/api/docs/models/gpt-5.6-luna APIAvailable via OpenAI API

API price $0.2/$1.2 · each benchmark row carries its own source badge (see methodology)